This work introduces a method that combines remote sensing and deep learning into a framework that is tailored for accurate, reliable and efficient counting and sizing of plants in aerial images. The investigated task focuses on two low-density crops, potato and lettuce. This double objective of counting and sizing is achieved through the detection and segmentation of individual plants by fine-tuning an existing deep learning architecture called Mask R-CNN. This paper includes a thorough discussion on the optimal parametrisation to adapt the Mask R-CNN architecture to this novel task. As we examine the correlation of the Mask R-CNN performance to the annotation volume and granularity (coarse or refined) of remotely sensed images of plants, we conclude that transfer learning can be effectively used to reduce the required amount of labelled data. Indeed, a previously trained Mask R-CNN on a low-density crop can improve performances after training on new crops. Once trained for a given crop, the Mask R-CNN solution is shown to outperform a manually-tuned computer vision algorithm. Model performances are assessed using intuitive metrics such as Mean Average Precision (mAP) from Intersection over Union (IoU) of the masks for individual plant segmentation and Multiple Object Tracking Accuracy (MOTA) for detection. The presented model reaches an mAP of 0.418 for potato plants and 0.660 for lettuces for the individual plant segmentation task. In detection, we obtain a MOTA of 0.781 for potato plants and 0.918 for lettuces.
Surface water availability is a fundamental environmental variable to implement effective climate adaptation and mitigation plans, as expressed by scientific, financial and political stakeholders. Recently published requirements urge the need for homogenised access to long historical records at a global scale, together with the standardised characterisation of the accuracy of observations. While satellite altimeters offer world coverage measurements, existing initiatives and online platforms provide derived water level data. However, these are sparse, particularly in complex topographies. This study introduces a new methodology in two steps (1) teroVIR, a virtual station extractor for a more comprehensive global and automatic monitoring of water bodies, and (2) teroWAT, a multi-mission, interoperable water level processor, for handling all terrain types. L2 and L1 altimetry products are used, with state-of-the-art retracker algorithms in the methodology. The work presents a benchmark between teroVIR and current platforms in West Africa, Kazakhastan and the Arctic: teroVIR shows an unprecedented increase from 55% to 99% in spatial coverage. A large-scale validation of teroWAT results in an average of unbiased root mean square error ubRMSE of 0.638 m on average for 36 locations in West Africa. Traditional metrics (ubRMSE, median, absolute deviation, Pearson coefficient) disclose significantly better values for teroWAT when compared with existing platforms, of the order of 8 cm and 5% improved respectively in error and correlation. teroWAT shows unprecedented excellent results in the Arctic, using an L1 products-based algorithm instead of L2, reducing the error by almost 4 m on average. To further compare teroWAT with existing methods, a new scoring option, teroSCO, is presented, measuring the quality of the validation of time series transversally and objectively across different strategies. Finally, teroVIR and teroWAT are implemented as platform-agnostic modules and used by flood forecasting and river discharge methods as relevant examples. A review of various applications for miscellaneous end-users is given, tackling the educational challenge raised by the community.
Surface water availability is a fundamental environmental variable to implement effective climate adaptation and mitigation plans, as expressed by scientific, financial and political stakeholders. Recently published requirements urge the need for homogenised access to long historical records at a global scale, together with the standardised characterisation of the accuracy of observations. While satellite altimeters offer world coverage measurements, existing initiatives and online platforms provide derived water level data. However, these are sparse, particularly in complex topographies. This study introduces a new methodology in two steps 1) teroVIR, a virtual station extractor for a more comprehensive global and automatic monitoring of water bodies, and 2) teroWAT, a multi-mission, interoperable water level processor, for handling all terrain types. L2 and L1 altimetry products are used, with state-of-the-art retracker algorithms in the methodology. The work presents a benchmark between teroVIR and current platforms in West Africa, Kazakhastan and the Arctic: teroVIR shows an unprecedented increase from 55% to 99% in spatial coverage.A large-scale validation of teroWAT results in an average of unbiased root mean square error ubRMSE of 0.638 m on average for 36 locations in West Africa. Traditional metrics (ubRMSE, median, absolute deviation, Pearson coefficient) disclose significantly better values for teroWAT when compared with existing platforms, of the order of 8 cm and 5% improved respectively in error and correlation. teroWAT shows unprecedented excellent results in the Arctic, using a L1 products based algorithm instead of L2 one, reducing the error of almost 4 m on average. To further compare teroWAT with existing methods, a new scoring option, teroSCO, is presented, measuring the quality of the validation of time series transversally and objectively across different strategies. Finally, teroVIR and teroWAT are implemented as platform-agnostic modules and used by flood forecasting and river discharge methods as relevant examples. A review of various applications for miscellaneous end-users is given, tackling the educational challenge raised by the community.
<p>Water level information is highly sought-after by operational hydrologists and emergency managers to improve flood management in, for example, West Africa (Lienert et al 2020). A main constraint of large-scale flood forecasting systems is an inability to convert streamflow volumes to water level at specific locations. Accurately representing water levels &#8211; and hence potential impacts of peak flows on local scale &#8211; is possible through detailed field work and hydraulic simulations (e.g. Massazza et al 2020). However, large-scale implementation of such approaches is typically constrained by lack of detailed topographic data. In this study we therefore develop a pragmatic method to estimate water levels using rating curves created through a combination of ground-based (in-situ) hydrometric gauge observations, hydrological simulations, and satellite altimetry data. Specifically, rating curves are created based on simulated discharge from HYPE models and 305 in-situ discharge observations from 1980 to 2020, in addition to 42 in-situ and 558 virtual water level stations (i.e. locations where Sentinel-3 missions intersect large rivers) from 2018 to 2020. The rating curves were estimated by fitting a conventional power-law equation. For the in-situ data this could be done directly from the two variables. These were, however, very scarce. We therefore exploited the EO-based virtual stations to be able to predict water levels at many more locations. To this end, rating curves were estimated using simulated discharge together with EO-based water level data at the virtual station locations. The inverted rating curve equation was subsequently used to transform simulated discharge to water level. The water levels estimated from simulated discharge were finally compared with the in-situ and virtual altimetry stations using accuracy performance metrics such as Nash-Sutcliffe efficiency (NSE) and Kling-Gupta efficiency (KGE). Furthermore, we examine and compare the rating curve uncertainty obtained from different data sources (in-situ, modelled and satellite data). This pragmatic methodology can be used in operational hydrology, specifically flood forecasting, to render forecasts more relevant at local scale and hence enable better flood risk management.</p>
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.